A knowledge-based genetic heuristic for learning certainty factors

Abstract

An expert network is a type of inference network that is derived from an expert system. One of the uses of expert networks is to to refine measures of certainty in knowledge bases using neural network learning techniques. Goal-directed Monte Carlo search (GDMC) is a parallel stochastic hillclimbing method that is being successfully used to refine certainty factors from data. This paper presents a new heuristic for GDMC that improves its performance by incorporating genetic algorithm techniques.<>